If you look at markets on crazy days and think “this feels like a story, not just numbers,” you’re already halfway to the idea behind the connection between narratives and realized volatility.
We’re going to unpack what that actually means, how to use it in practice, and what to watch out for in 2025.
—
What do мы mean by “narratives” and “realized volatility”?
In plain language, a market *narrative* is the shared story people tell each other about what’s going on:
– “AI will eat every industry.”
– “The Fed is done hiking; soft landing is coming.”
– “Geopolitics will break supply chains again.”
These stories spread through media, social networks, research notes and chats between traders. Narrative economics and financial markets are glued together by attention: what people talk about shapes what they trade and how nervous they feel about the future.
Realized volatility is the measured choppiness of prices over a period, usually built from high‑frequency intraday returns. It’s not “guessed,” it’s *observed*:
– You take, say, 5‑minute returns over the trading day.
– Square them, sum them, annualize: that’s realized vol.
Realized volatility data for quantitative trading is the backbone for risk management, option pricing, and performance attribution. It tells you how violent the actual path of prices has been, not just what the VIX or implied volatility thought beforehand.
The diagnostic question is: how much of that actual choppiness is driven by stories, not just by the usual macro and micro factors?
—
Three main ways to connect stories and volatility
In practice, there are three broad approaches people use today to link narratives with realized volatility.
1. Calendar-based “event windows”
2. Text and sentiment models
3. Full‑blown narrative indices
Let’s walk through them in practical terms.
—
1. Event windows: the simple but surprisingly effective workhorse
This is the oldest and most transparent method. You take dates of big narrative events — Fed meetings, CPI releases, earnings, elections, wars, regulatory shocks — and look at realized volatility before and after.
Concretely:
1. Define an event (e.g., Fed announcement).
2. Compute realized volatility on:
– a pre‑event window (say, −5 to −1 days),
– the event day (0),
– a post‑event window (+1 to +5 days).
3. Compare the patterns over many events.
This directly tackles the economic news impact on stock market volatility: are CPI days structurally “louder” than normal days? Does volatility spike *before* earnings (anticipation) or *after* (surprise)?
Pros (why people still do this):
– Clear and intuitive: PMs, regulators and risk committees understand “CPI day vol is 3× normal.”
– Easy to implement: needs only a reliable event calendar and intraday price data.
– Great for diagnostics: good at answering “what happened around X?”
Cons (what you’ll hit quickly):
– Narratives rarely fit cleanly into one day; they build and decay over weeks.
– Overlapping stories: a Fed meeting might land during a geopolitical shock.
– You only see volatility spikes, not the *content* or tone of the conversation.
In practice, this is a baseline. Many firms still start their realized volatility forecasting models from these simple patterns and then layer more sophisticated narrative data on top.
—
2. Text, sentiment, and “what are people actually saying?”
The next layer is to stop guessing what the story is and actually read what’s being said — at scale.
This is where NLP (natural language processing) enters:
– News headlines and articles
– Earnings call transcripts
– Social media and forums
– Broker research and blogs
The workflow usually looks like this:
1. Collect text time‑series aligned with your assets or indices.
2. Clean and normalize the text (entities, tickers, languages).
3. Use models to tag sentiment (bullish/bearish), uncertainty, topics (inflation, AI, war, regulation).
4. Turn these into daily or intraday indices.
You then check which narrative features help explain or predict realized volatility. For example, you may find that spikes in “policy uncertainty” language lead to higher intraday vol over the next two days in interest‑rate futures.
Strengths:
– Much richer than “event/no event” — you see *gradual* narrative build‑ups.
– You can distinguish types of fear: macro uncertainty vs. firm‑specific bad news.
– Backtests can be very granular at the sector or single‑name level.
Weaknesses:
– Data cleaning is painful and never really “finished.”
– Model drift: language changes fast; models trained on 2018 news may misread memes in 2025.
– Risk of overfitting: with so many text features, it’s easy to find patterns that don’t generalize.
This is the core of modern narrative economics and financial markets research: treating words as data and pairing them with realized volatility instead of just prices or returns.
—
3. Narrative indices: from scattered stories to standardized factors
The most advanced camp builds aggregate “narrative factors” and feeds them directly into forecasting and trading systems.
Think of things like:
– A global “inflation story” index
– A “AI productivity boom” theme index
– A “geopolitical risk” index
The idea is to compress the messy world of text into a handful of daily time‑series. These sit alongside your usual macro factors and are plugged into realized volatility forecasting models or risk systems.
Typical use:
– Use topic models or embeddings to cluster text into coherent storylines.
– Estimate how strongly each storyline is “active” each day.
– Regress realized volatility on these narrative indices, along with usual controls (past vol, macro releases, liquidity indicators).
This gives you a diagnostic lens: not just “vol is up,” but “vol is up mainly because the ‘policy uncertainty’ narrative is dominating.”
Upsides:
– Extremely powerful for storytelling to stakeholders: “Here’s which themes drove risk this quarter.”
– Good for scenario analysis: “What if the ‘trade war’ narrative reactivates to 2022 levels?”
– Scales well across asset classes once the pipeline is in place.
Downsides:
– Methodologically heavy — needs strong quant and data engineering teams.
– Transparency issues: some topic models produce “weird” factors that are hard to interpret.
– The link from narrative factor → actionable trade signal can be fuzzy if not designed carefully.
—
How does this change real trading decisions?
Let’s get concrete. For volatility trading strategies for institutional investors, the link between narratives and realized volatility matters in three main ways.
1. Timing of exposure
You might run systematic strategies selling options when realized vol is expected to stay low relative to implied, or buying options ahead of expected spikes.
If your models detect a strong, fast‑moving narrative about policy shock or systemic risk, you can:
– Reduce short‑vol positions before the market “wakes up.”
– Switch from short gamma to long gamma around expected narrative peaks.
– Change tenor: prefer shorter‑dated options when narratives are explosive but fleeting.
2. Cross‑section: what to trade, not just when
Different narratives hit different assets. An “AI productivity” narrative might pump tech single names’ realized volatility long before it significantly moves broad indices.
Linking thematic narrative indices to realized vol helps you:
– Choose which sectors to run relative‑value vol trades on.
– Allocate risk in a dispersion strategy: overweight names with story‑driven vol.
– Avoid being short vol in the precise pockets the narrative is likely to set on fire.
3. Risk and sizing
For many institutional investors, the main use is not alpha but *position sizing and drawdown control*:
– If your narrative indicators say “high risk of regime change,” you can enforce higher vol floors.
– Scenario tests: “What happens to our portfolio vol if the ‘recession’ narrative spikes like in 2020?”
In short, narratives become an extra risk factor that modulates how you translate model output into trading size.
—
Comparing approaches: when is “fancy” actually better?
If you strip away the technical buzzwords, you’re often choosing between:
1. Simple calendar‑events‑plus‑historical‑patterns.
2. Medium‑complexity sentiment and topic indices from text.
3. High‑complexity, fully integrated narrative‑volatility systems.
Here’s a practical way to compare them:
1. Data costs and engineering
2. Transparency to decision‑makers
3. Speed to adapt in new crises
4. Real impact on P&L or risk metrics
As a rule of thumb:
– If your horizon is days–weeks and you trade liquid index options, event‑based and light sentiment models get you 70–80% of the benefit with 20% of the complexity.
– If you’re doing single‑name or sector volatility relative value, more granular text‑based narratives pay off because micro‑stories matter a lot.
– If you run a large, multi‑asset book with regulatory oversight, narrative indices help with *explainability*: you can tell supervisors and clients *why* volatility changed, not just that it did.
—
Pros and cons of the underlying technologies
Let’s zoom in on the tech layer itself.
What works well:
– High‑frequency realized vol estimation: The math is solid and robust, especially with liquid instruments. For most assets, you can get reliable realized vol measures intraday.
– Embedding‑based NLP models: Modern language models capture subtle shifts in tone and context better than old‑school bag‑of‑words or naive sentiment scores.
– Cloud pipelines: Cheaper and more scalable than five years ago; running daily narrative indices is now accessible to mid‑sized funds.
Where the landmines are:
– Noise from social media: The loudest narrative isn’t always the one that drives capital flows. You need filters (by user type, reach, or historical relevance).
– Look‑ahead and survivorship bias: When you build narrative features, it’s easy to accidentally use information that wasn’t visible at the time.
– Interpretability vs. accuracy trade‑off: The most accurate models are often black boxes. For risk and compliance, “why” can be more important than squeezing another 2% out of forecast error.
For many desks, the best compromise is a hybrid: transparent features (event flags, simple sentiment scores) combined with a small set of more complex narrative indices whose behavior they monitor closely.
—
Recommendations: how to start in practice
If you’re thinking about bringing narrative–volatility diagnostics into your workflow, here’s a pragmatic roadmap:
1. Nail the basics of realized volatility first
– Make sure your intraday data is clean and your realized vol estimates are stable.
– Benchmark simple autoregressive or HAR‑type realized volatility forecasting models — you need a baseline to judge whether narratives add value.
2. Layer in events and simple narrative proxies
– Start with an annotated calendar (macro releases, central bank, earnings, elections).
– Add a few hand‑crafted indices like “policy uncertainty” or “macro surprise” based on curated news or survey data.
– Check: do these improve out‑of‑sample forecasts or risk metrics?
3. Introduce text‑based signals where they matter most
– Pick 1–2 asset classes where you *know* stories drive behavior (e.g., tech growth stocks, EM FX, crypto‑linked equities).
– Build modest NLP features: sentiment on a limited set of sources, or simple topic tags.
– Use them to adjust risk, not to overhaul your entire trading engine on day one.
4. Move to full narrative indices only with a clear use case
– Don’t build narrative indices “for fun.” Tie each index to:
– A specific trading decision (e.g., sizing of short‑vol trades in rates).
– A specific risk question (e.g., exposure to “geopolitical shock” scenarios).
– Document how and when these indices should override or modify your default vol forecasts.
This incremental approach also makes it easier to convince risk committees and investors: they can see step‑by‑step where each layer adds explanatory or predictive power.
—
Trends to watch in 2025
Looking forward, a few trends are shaping how narratives and realized volatility will be used:
– Intraday narrative signals: Instead of daily indices, more desks are experimenting with near‑real‑time feeds — streaming news and social data that adjust intraday vol forecasts during the session.
– Cross‑asset narrative mapping: The same story — “AI boom,” “re‑shoring,” “China slowdown” — affects equities, rates, credit, commodities, FX differently. 2025 tools increasingly track how a single theme propagates through the whole complex.
– Regulatory attention: Supervisors are becoming more interested in models that rely on alternative data and AI. Being able to show a clear, auditable link from narrative inputs to vol forecasts is becoming a competitive advantage, not a burden.
– Human–machine collaboration: Discretionary PMs and systematic teams are finally meeting in the middle. Narrative dashboards inform human judgment, while realized volatility models keep the numbers honest. Expect more desks where fundamental and quant teams share the same narrative analytics tools.
The big picture: we’re moving from “stories as color” and “volatility as pure math” to a shared framework where the two talk to each other systematically.
—
Bottom line: using stories without getting lost in them
Narratives won’t magically turn a bad strategy into a good one, and no amount of clever NLP will remove the need for solid risk discipline. But if you already rely on realized vol to manage risk or run options books, ignoring the narrative side is leaving information on the table.
Treat narratives as a measurable risk factor, not as folklore. Start simple, validate against your own realized volatility patterns, and only then graduate to the fancy stuff. In 2025, the edge is less about having the flashiest model and more about using the connection between stories and volatility in a controlled, testable, and explainable way.